Self-driving vehicle programming

[jcl]

Originally Posted by richard5933

Either way - platooning or neural net, they both rely on the first car to approach the obstacle recognizing it and taking the correct evasive action. If the first car misses it, so will the ones that follow.

Regardless, I foresee this as becoming the fastest-growing area of legal practice in the coming years. I can just see the late night headlines now, "Were you or someone you loved injured or killed as a result of an missprogrammed self-driving vehicle? If so..."

Platooning means that the following vehicle does the same as the vehicle in the lead.

The neural net is an AI learning system. The vehicle that avoids a pothole can be the only one on the road, not following anything. The information about the road hazard comes from many previous vehicles, and camera data, not a lead vehicle.

[richard5933]

Originally Posted by jcl

Platooning means that the following vehicle does the same as the vehicle in the lead.

The neural net is an AI learning system. The vehicle that avoids a pothole can be the only one on the road, not following anything. The information about the road hazard comes from many previous vehicles, and camera data, not a lead vehicle.

I know that the words mean. Perhaps I just didn't make my point clear.

Whether its by platooning or by learning through a neural net, a self-driving vehicle still needs to learn that there's an obstacle. Your earlier post mentioned learning from "many vehicles steer around the pothole or avoid the crumbling edge of the roadway", which is great if many vehicles in the neural net have previously avoided the obstacle.

Not sure what road conditions are like by you, but around here (especially in the winter & early spring) road hazards literally can occur from one hour to the next. Anything from an entire section of asphalt heaving up to a chuck of asphalt popping out overnight during a deep freeze. Someone has to be the first car to approach these obstacles, and often time they are very difficult to discern from a wet spot or another benign road condition.

Since we're not that far removed from self-driving cars trying to swerve to avoid snow flakes, and with my experience of AI pushing a car to drive through broken pavement, it doesn't seem that AI is quite at the level of being able to make the same kind of fine distinctions a human eye can make. At least not yet.

[jcl]

Originally Posted by richard5933

I know that the words mean. Perhaps I just didn't make my point clear.

Whether its by platooning or by learning through a neural net, a self-driving vehicle still needs to learn that there's an obstacle. Your earlier post mentioned learning from "many vehicles steer around the pothole or avoid the crumbling edge of the roadway", which is great if many vehicles in the neural net have previously avoided the obstacle.

Not sure what road conditions are like by you, but around here (especially in the winter & early spring) road hazards literally can occur from one hour to the next. Anything from an entire section of asphalt heaving up to a chuck of asphalt popping out overnight during a deep freeze. Someone has to be the first car to approach these obstacles, and often time they are very difficult to discern from a wet spot or another benign road condition.

Since we're not that far removed from self-driving cars trying to swerve to avoid snow flakes, and with my experience of AI pushing a car to drive through broken pavement, it doesn't seem that AI is quite at the level of being able to make the same kind of fine distinctions a human eye can make. At least not yet.

It doesn't just rely on the neural net to avoid obstacles. It relies on cameras on the vehicle. The neural net is an enhancement that improves performance, not the primary navigation system.

Road conditions near me mean that in winter, especially on roads that are fully or partially snow covered, Autopilot disengages often enough that I don't bother using it at times. But even when not engaged, the active safety features will provide an alert if I get too close to the edge of the road, and that still provides a safety benefit. The dash screen regularly shows the road lines from the very sensitive cameras (8 or more) better than I can see with my eyes.

As to whether these systems can perform at the level of a human operator, we need to consider the crash data reported every quarter. They are already performing significantly better than human drivers. Not at every moment and on every road condition, but if I was in a crash and the other vehicle had their Autopilot or equivalent system turned off, I would categorize that as contributing. I don't know if contributory negligence is the correct legal term, but that idea.

[richard5933]

Originally Posted by jcl

It doesn't just rely on the neural net to avoid obstacles. It relies on cameras on the vehicle. The neural net is an enhancement that improves performance, not the primary navigation system.

Road conditions near me mean that in winter, especially on roads that are fully or partially snow covered, Autopilot disengages often enough that I don't bother using it at times. But even when not engaged, the active safety features will provide an alert if I get too close to the edge of the road, and that still provides a safety benefit. The dash screen regularly shows the road lines from the very sensitive cameras (8 or more) better than I can see with my eyes.

As to whether these systems can perform at the level of a human operator, we need to consider the crash data reported every quarter. They are already performing significantly better than human drivers. Not at every moment and on every road condition, but if I was in a crash and the other vehicle had their Autopilot or equivalent system turned off, I would categorize that as contributing. I don't know if contributory negligence is the correct legal term, but that idea.

But what happens when the autopilot is engaged and there is a collision? Who's liable then?

[jcl]

Originally Posted by richard5933

But what happens when the autopilot is engaged and there is a collision? Who's liable then?

Depends. First of all, it may be the other vehicle, and not the AI vehicle. It can help that multiple cameras are documenting everything, but for purposes of your question, let's assume that the other vehicle was not at fault.

With Tesla, the driver remains responsible, and that is made very clear in their documentation and literature. In fact, if the driver isn't paying attention they turn off the system to stress that point.

With Mercedes, as a recent example, the manufacturer. Because they are promoting their Level 3 system. But there are lots of conditions. No night driving, only specific roads, not over 40 mph, no construction zones, no inclement weather, no tunnels, no emergency vehicles on the road, etc.

https://cleantechnica.com/2022/03/23...em-is-engaged/

[skyguyscott]

Twenty-five years ago, I was working at a former subsidiary of a major automobile manufacturer, and already, they were working on self-driving vehicles. One approach they were developing at the time involved embedding lip-stick-sized magnets in the pavement.

But even back then, I was struck by the potential liability issues inherent in offering the public a self-driving vehicle. Knowing the disposition of corporate legal departments as I have come to, I was certain no major company would ever, ever field a self-driving car.

So I have been shocked to see how, somehow, someway, congress has been persuaded to shield automakers from liability suits in the development of autonomous vehicles, at least, for now.

I don't know if, way back then, the engineers had thought far enough ahead about how to program the vehicle to respond to various scenarios, or especially about how tightly or loosely to hew to various traffic laws, as opposed to just getting the system to work, but here we are today, and the technology is still so new, a consensus has yet to emerge.

[Authority924]

Those cars also take in sight the road marking, the lanes and the stop lanes, also takes in sight the road signs, it's well known it will take years and years to learn them to drive without any help from the driver, but the technology is still a little bit raw